Abdominal Aorta Segmentation
Abdominal Aorta and Anatomical Models
The abdominal aorta is usually located to the left of the center of the body and runs along the spine toward the lower abdomen, where it bifurcates. Anatomically important arteries, such as the celiac and superior mesenteric arteries arise from the abdominal aorta. Therefore, abdominal aortic segmentation is essential to recognize the network of abdominal arteries.
It is easy to segment the abdominal aorta region from a contrast-enhanced CT volume of a patient who does not suffer from aortic diseases such an aneurysm, since the contour of the aorta is very clear. A simple region-growing method can achieve sufficient segmentation. However, for CT volumes from a patient suffering from an abdominal aortic aneurysm, or non-contrast-enhanced CT volumes, the segmentation of the aorta focusing only on the intensity distribution does not work well. It causes the lack of the aneurysm area or over-segmentation of other surrounding organs. Model-based segmentation methods have been reported [46, 73, 157, 227, 334] to cope with these difficulties.
Since the aorta is a tubular organ, the aorta model consists of a tube surface. A centerline of the aorta is also modeled together with a tube surface. The tube surface is constructed by parametric surfaces such as B-spline. A surface model may also have the appearance of the aortic contour which represents the intensity or edge distribution at the vicinity of the contour. The distribution information is analyzed using PCA, and the appearance is often represented as eigenvectors.
Abdominal Aorta Segmentation The region-growing approach to segmentation becomes inaccurate in the face of abnormalities such as aortic aneurysms. There are several approaches to maintain segmentation accuracy.
The Centerline Model and Hough Transformation First, the centerline model of the aorta is fitted to an input CT volume. One can make a likelihood map of the aortic centerline by detecting edges of the aorta and performing the distance transformation to the edge voxels. The central parts of the aorta tend to have larger distance, and the distance information is used as likelihoods of the aortic centerline. A model fitting technique is applied which maximizes likelihoods on the centerline model. Then, the aortic surface is recovered by performing the reverse distance transformation using distance values on the centerline. In the recovery step, the Hough transformation may be adopted for modification of the distance value. If false edges are detected in the construction of a likelihood map, the distance value on the centerline becomes smaller. The Hough transformation is thus applied to the edge voxels, and the center points of the aorta are modified to the location which has the maximum votes.
The Tube Surface Model The tube surface model is used to segment the aortic surface more accurately. After the centerline model of the aorta is fitted, the surface model is deformed so as to fit it to the contour of the aorta. The likelihood map can be used as the deforming energy. The appearance of the aortic contour can also be added to the energy.